Use ExecutionStatus in learning and building an evaluation dataset

[jlewi]

#310 should start logging execution status. Once we have that data we should take into account when learning.
If a code cell didn't execute correctly we probably shouldn't learn from it because the command is wrong.

Similarly in DumpExample we should probably filter out examples created from executions that didn't succeed.